Marmotize

AI engineer with Imposter Syndrome

[Marmotize @marmotize_]

@Jonny9811 @skelz0r @JoshARosen can you explain more how and where the decision model fits?

[Marmotize @marmotize_]

@skelz0r @JoshARosen adding visibility on*

[Marmotize @marmotize_]

@skelz0r @JoshARosen We are almost there banning comments but we still value some and adding visibility and comment bloat is helping.

[Marmotize @marmotize_]

@JoshARosen its all deterministic scripts from the code, doing dependency analysis with dependency-cruiser for example

[Marmotize @marmotize_]

We have a Github action in my team that surfaces those metadata information about a PR and posts a comment on the PR with labels to help reviewers with their load and focus on what they care about mostly:

- new external dependencies added
- new links established between internal components
- number of comments added and removed
- edits to the AGENTS.md / arch docs

[Marmotize @marmotize_]

@FUCORY I agree and thats why I have been using ephermal workspaces lately with a TTL per task and I dont rely on memory but actual skills if needed (not much lately)

[Max Stoiber @mxstbr]

there's a nascent trend of hiding MCP tools behind two generic tools: search & execute.

i don't think it's a good default. this basically means implementing tool search & code mode on each MCP server.

but clients like codex already have tool search & code mode that a/ work across all MCP servers and b/ we train our models to be great at!

why make every MCP server build that layer and then make the model navigate both, including a custom one on each MCP server that it isn't trained to be great at?!

expose your actual tools and let clients* find and compose them.

*clients that support tool search & code mode like ChatGPT and Codex

[Marmotize @marmotize_]

@DominikTornow @resonatehqio Hey @DominikTornow, Big fan of your work and I am actually working on something very similar in my company and using AWS Lambda Durable functions, I am wondering though why did you choose QuickJS?

[Claude @claudeai]

We're extending access to Claude Fable 5 on all paid plans through July 12.

[Thariq @trq212]

https://t.co/hPiZr1kG7r

[Marmotize @marmotize_]

@realhenry @gregisenberg Hey @realhenry, check https://t.co/yB6UICeEDG

unless you dont use xurl and have some browser automation 🫣

[Marmotize @marmotize_]

Cloudflare is the AWS of the future.

[Marmotize @marmotize_]

@plannotator This is super cool! Very similar to a project I was working on! Will be looking at it soon!

[Guinness Chen @guinnesschen]

If you ever get tired of managing your Codex threads, just let Codex manage itself! Codex can now create threads, search them, organize them, pin the important ones, and spin up worktrees for parallel tasks.

[Marmotize @marmotize_]

@steipete love the local first direction, I am yet to find time to work on a chatgpt chats archive (hope this is tos ok 🫣)

[Marmotize @marmotize_]

@FUCORY @a_protsyuk Thank you! Smithers is already on our radar 😉. We deal with tens of millions of workflow executions per day so we had to write our own for cost reasons but I am always scouting for options. Also cohorts for fleets and other advanced topics are a must for us.

[Marmotize @marmotize_]

@pvncher @RepoPrompt @karpathy Cool! I will integrate that with https://t.co/llN10v56wd

[Marmotize @marmotize_]

@steipete Thanks @steipete this will be very useful. Do you think a similar tool but for storing locally our ChatGPT chats would also benefecial? I was thinking of trying that out.